4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License version 2 for more details. A copy is
14 * included in the COPYING file that accompanied this code.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Copyright 2009 Sun Microsystems, Inc. All rights reserved
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2013, Intel Corporation.
29 * lustre/lod/lod_object.c
31 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
34 #define DEBUG_SUBSYSTEM S_MDS
37 #include <obd_class.h>
38 #include <lustre_ver.h>
39 #include <obd_support.h>
40 #include <lprocfs_status.h>
42 #include <lustre_fid.h>
43 #include <lustre_param.h>
44 #include <lustre_fid.h>
45 #include <lustre_lmv.h>
47 #include <md_object.h>
49 #include "lod_internal.h"
51 static const char dot[] = ".";
52 static const char dotdot[] = "..";
54 extern struct kmem_cache *lod_object_kmem;
55 static const struct dt_body_operations lod_body_lnk_ops;
57 static int lod_index_lookup(const struct lu_env *env, struct dt_object *dt,
58 struct dt_rec *rec, const struct dt_key *key,
59 struct lustre_capa *capa)
61 struct dt_object *next = dt_object_child(dt);
62 return next->do_index_ops->dio_lookup(env, next, rec, key, capa);
65 static int lod_declare_index_insert(const struct lu_env *env,
67 const struct dt_rec *rec,
68 const struct dt_key *key,
69 struct thandle *handle)
71 return dt_declare_insert(env, dt_object_child(dt), rec, key, handle);
74 static int lod_index_insert(const struct lu_env *env,
76 const struct dt_rec *rec,
77 const struct dt_key *key,
79 struct lustre_capa *capa,
82 return dt_insert(env, dt_object_child(dt), rec, key, th, capa, ign);
85 static int lod_declare_index_delete(const struct lu_env *env,
87 const struct dt_key *key,
90 return dt_declare_delete(env, dt_object_child(dt), key, th);
93 static int lod_index_delete(const struct lu_env *env,
95 const struct dt_key *key,
97 struct lustre_capa *capa)
99 return dt_delete(env, dt_object_child(dt), key, th, capa);
102 static struct dt_it *lod_it_init(const struct lu_env *env,
103 struct dt_object *dt, __u32 attr,
104 struct lustre_capa *capa)
106 struct dt_object *next = dt_object_child(dt);
107 struct lod_it *it = &lod_env_info(env)->lti_it;
108 struct dt_it *it_next;
111 it_next = next->do_index_ops->dio_it.init(env, next, attr, capa);
115 /* currently we do not use more than one iterator per thread
116 * so we store it in thread info. if at some point we need
117 * more active iterators in a single thread, we can allocate
119 LASSERT(it->lit_obj == NULL);
121 it->lit_it = it_next;
124 return (struct dt_it *)it;
127 #define LOD_CHECK_IT(env, it) \
129 LASSERT((it)->lit_obj != NULL); \
130 LASSERT((it)->lit_it != NULL); \
133 void lod_it_fini(const struct lu_env *env, struct dt_it *di)
135 struct lod_it *it = (struct lod_it *)di;
137 LOD_CHECK_IT(env, it);
138 it->lit_obj->do_index_ops->dio_it.fini(env, it->lit_it);
140 /* the iterator not in use any more */
145 int lod_it_get(const struct lu_env *env, struct dt_it *di,
146 const struct dt_key *key)
148 const struct lod_it *it = (const struct lod_it *)di;
150 LOD_CHECK_IT(env, it);
151 return it->lit_obj->do_index_ops->dio_it.get(env, it->lit_it, key);
154 void lod_it_put(const struct lu_env *env, struct dt_it *di)
156 struct lod_it *it = (struct lod_it *)di;
158 LOD_CHECK_IT(env, it);
159 return it->lit_obj->do_index_ops->dio_it.put(env, it->lit_it);
162 int lod_it_next(const struct lu_env *env, struct dt_it *di)
164 struct lod_it *it = (struct lod_it *)di;
166 LOD_CHECK_IT(env, it);
167 return it->lit_obj->do_index_ops->dio_it.next(env, it->lit_it);
170 struct dt_key *lod_it_key(const struct lu_env *env, const struct dt_it *di)
172 const struct lod_it *it = (const struct lod_it *)di;
174 LOD_CHECK_IT(env, it);
175 return it->lit_obj->do_index_ops->dio_it.key(env, it->lit_it);
178 int lod_it_key_size(const struct lu_env *env, const struct dt_it *di)
180 struct lod_it *it = (struct lod_it *)di;
182 LOD_CHECK_IT(env, it);
183 return it->lit_obj->do_index_ops->dio_it.key_size(env, it->lit_it);
186 int lod_it_rec(const struct lu_env *env, const struct dt_it *di,
187 struct dt_rec *rec, __u32 attr)
189 const struct lod_it *it = (const struct lod_it *)di;
191 LOD_CHECK_IT(env, it);
192 return it->lit_obj->do_index_ops->dio_it.rec(env, it->lit_it, rec, attr);
195 __u64 lod_it_store(const struct lu_env *env, const struct dt_it *di)
197 const struct lod_it *it = (const struct lod_it *)di;
199 LOD_CHECK_IT(env, it);
200 return it->lit_obj->do_index_ops->dio_it.store(env, it->lit_it);
203 int lod_it_load(const struct lu_env *env, const struct dt_it *di, __u64 hash)
205 const struct lod_it *it = (const struct lod_it *)di;
207 LOD_CHECK_IT(env, it);
208 return it->lit_obj->do_index_ops->dio_it.load(env, it->lit_it, hash);
211 int lod_it_key_rec(const struct lu_env *env, const struct dt_it *di,
214 const struct lod_it *it = (const struct lod_it *)di;
216 LOD_CHECK_IT(env, it);
217 return it->lit_obj->do_index_ops->dio_it.key_rec(env, it->lit_it, key_rec);
220 static struct dt_index_operations lod_index_ops = {
221 .dio_lookup = lod_index_lookup,
222 .dio_declare_insert = lod_declare_index_insert,
223 .dio_insert = lod_index_insert,
224 .dio_declare_delete = lod_declare_index_delete,
225 .dio_delete = lod_index_delete,
233 .key_size = lod_it_key_size,
235 .store = lod_it_store,
237 .key_rec = lod_it_key_rec,
241 static void lod_object_read_lock(const struct lu_env *env,
242 struct dt_object *dt, unsigned role)
244 dt_read_lock(env, dt_object_child(dt), role);
247 static void lod_object_write_lock(const struct lu_env *env,
248 struct dt_object *dt, unsigned role)
250 dt_write_lock(env, dt_object_child(dt), role);
253 static void lod_object_read_unlock(const struct lu_env *env,
254 struct dt_object *dt)
256 dt_read_unlock(env, dt_object_child(dt));
259 static void lod_object_write_unlock(const struct lu_env *env,
260 struct dt_object *dt)
262 dt_write_unlock(env, dt_object_child(dt));
265 static int lod_object_write_locked(const struct lu_env *env,
266 struct dt_object *dt)
268 return dt_write_locked(env, dt_object_child(dt));
271 static int lod_attr_get(const struct lu_env *env,
272 struct dt_object *dt,
273 struct lu_attr *attr,
274 struct lustre_capa *capa)
276 return dt_attr_get(env, dt_object_child(dt), attr, capa);
279 static int lod_declare_attr_set(const struct lu_env *env,
280 struct dt_object *dt,
281 const struct lu_attr *attr,
282 struct thandle *handle)
284 struct dt_object *next = dt_object_child(dt);
285 struct lod_object *lo = lod_dt_obj(dt);
290 * declare setattr on the local object
292 rc = dt_declare_attr_set(env, next, attr, handle);
296 /* osp_declare_attr_set() ignores all attributes other than
297 * UID, GID, and size, and osp_attr_set() ignores all but UID
298 * and GID. Declaration of size attr setting happens through
299 * lod_declare_init_size(), and not through this function.
300 * Therefore we need not load striping unless ownership is
301 * changing. This should save memory and (we hope) speed up
303 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
304 if (!(attr->la_valid & (LA_UID | LA_GID)))
307 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
310 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE |
311 LA_ATIME | LA_MTIME | LA_CTIME)))
315 * load striping information, notice we don't do this when object
316 * is being initialized as we don't need this information till
317 * few specific cases like destroy, chown
319 rc = lod_load_striping(env, lo);
323 if (lo->ldo_stripenr == 0)
326 if (!(attr->la_valid & ~(LA_ATIME | LA_MTIME | LA_CTIME))) {
327 struct lu_attr *la = &lod_env_info(env)->lti_attr;
328 bool setattr_time = false;
330 rc = dt_attr_get(env, dt_object_child(dt), la,
335 /* If it will only setattr time, it will only set
336 * time < current_time */
337 if ((attr->la_valid & LA_ATIME &&
338 attr->la_atime < la->la_atime) ||
339 (attr->la_valid & LA_CTIME &&
340 attr->la_ctime < la->la_ctime) ||
341 (attr->la_valid & LA_MTIME &&
342 attr->la_mtime < la->la_mtime))
349 * if object is striped declare changes on the stripes
351 LASSERT(lo->ldo_stripe);
352 for (i = 0; i < lo->ldo_stripenr; i++) {
353 LASSERT(lo->ldo_stripe[i]);
355 rc = dt_declare_attr_set(env, lo->ldo_stripe[i], attr, handle);
357 CERROR("failed declaration: %d\n", rc);
365 static int lod_attr_set(const struct lu_env *env,
366 struct dt_object *dt,
367 const struct lu_attr *attr,
368 struct thandle *handle,
369 struct lustre_capa *capa)
371 struct dt_object *next = dt_object_child(dt);
372 struct lod_object *lo = lod_dt_obj(dt);
377 * apply changes to the local object
379 rc = dt_attr_set(env, next, attr, handle, capa);
383 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr)) {
384 if (!(attr->la_valid & (LA_UID | LA_GID)))
387 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_OWNER))
390 if (!(attr->la_valid & (LA_UID | LA_GID | LA_MODE |
391 LA_ATIME | LA_MTIME | LA_CTIME)))
395 if (lo->ldo_stripenr == 0)
398 if (!(attr->la_valid & ~(LA_ATIME | LA_MTIME | LA_CTIME))) {
399 struct lu_attr *la = &lod_env_info(env)->lti_attr;
400 bool setattr_time = false;
402 rc = dt_attr_get(env, dt_object_child(dt), la,
407 /* If it will only setattr time, it will only set
408 * time < current_time */
409 if ((attr->la_valid & LA_ATIME &&
410 attr->la_atime < la->la_atime) ||
411 (attr->la_valid & LA_CTIME &&
412 attr->la_ctime < la->la_ctime) ||
413 (attr->la_valid & LA_MTIME &&
414 attr->la_mtime < la->la_mtime))
422 * if object is striped, apply changes to all the stripes
424 LASSERT(lo->ldo_stripe);
425 for (i = 0; i < lo->ldo_stripenr; i++) {
426 LASSERT(lo->ldo_stripe[i]);
427 rc = dt_attr_set(env, lo->ldo_stripe[i], attr, handle, capa);
429 CERROR("failed declaration: %d\n", rc);
437 static int lod_xattr_get(const struct lu_env *env, struct dt_object *dt,
438 struct lu_buf *buf, const char *name,
439 struct lustre_capa *capa)
441 struct lod_thread_info *info = lod_env_info(env);
442 struct lod_device *dev = lu2lod_dev(dt->do_lu.lo_dev);
446 rc = dt_xattr_get(env, dt_object_child(dt), buf, name, capa);
447 if (rc != -ENODATA || !S_ISDIR(dt->do_lu.lo_header->loh_attr & S_IFMT))
451 * lod returns default striping on the real root of the device
452 * this is like the root stores default striping for the whole
453 * filesystem. historically we've been using a different approach
454 * and store it in the config.
456 dt_root_get(env, dev->lod_child, &info->lti_fid);
457 is_root = lu_fid_eq(&info->lti_fid, lu_object_fid(&dt->do_lu));
459 if (is_root && strcmp(XATTR_NAME_LOV, name) == 0) {
460 struct lov_user_md *lum = buf->lb_buf;
461 struct lov_desc *desc = &dev->lod_desc;
463 if (buf->lb_buf == NULL) {
465 } else if (buf->lb_len >= sizeof(*lum)) {
466 lum->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V1);
467 lmm_oi_set_seq(&lum->lmm_oi, FID_SEQ_LOV_DEFAULT);
468 lmm_oi_set_id(&lum->lmm_oi, 0);
469 lmm_oi_cpu_to_le(&lum->lmm_oi, &lum->lmm_oi);
470 lum->lmm_pattern = cpu_to_le32(desc->ld_pattern);
471 lum->lmm_stripe_size = cpu_to_le32(
472 desc->ld_default_stripe_size);
473 lum->lmm_stripe_count = cpu_to_le16(
474 desc->ld_default_stripe_count);
475 lum->lmm_stripe_offset = cpu_to_le16(
476 desc->ld_default_stripe_offset);
486 static int lod_verify_md_striping(struct lod_device *lod,
487 const struct lmv_user_md_v1 *lum)
492 if (unlikely(le32_to_cpu(lum->lum_magic) != LMV_USER_MAGIC))
493 GOTO(out, rc = -EINVAL);
495 if (unlikely(le32_to_cpu(lum->lum_stripe_count) == 0))
496 GOTO(out, rc = -EINVAL);
498 if (unlikely(le32_to_cpu(lum->lum_stripe_count) >
499 lod->lod_remote_mdt_count + 1))
500 GOTO(out, rc = -EINVAL);
503 CERROR("%s: invalid lmv_user_md: magic = %x, "
504 "stripe_offset = %d, stripe_count = %u: rc = %d\n",
505 lod2obd(lod)->obd_name, le32_to_cpu(lum->lum_magic),
506 (int)le32_to_cpu(lum->lum_stripe_offset),
507 le32_to_cpu(lum->lum_stripe_count), rc);
511 int lod_prep_lmv_md(const struct lu_env *env, struct dt_object *dt,
512 struct lu_buf *lmv_buf)
514 struct lod_thread_info *info = lod_env_info(env);
515 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
516 struct lod_object *lo = lod_dt_obj(dt);
517 struct lmv_mds_md_v1 *lmm1;
520 int type = LU_SEQ_RANGE_ANY;
526 LASSERT(lo->ldo_dir_striped != 0);
527 LASSERT(lo->ldo_stripenr > 0);
528 stripe_count = lo->ldo_stripenr + 1;
529 lmm_size = lmv_mds_md_size(stripe_count, LMV_MAGIC);
530 if (info->lti_ea_store_size < lmm_size) {
531 rc = lod_ea_store_resize(info, lmm_size);
536 lmm1 = (struct lmv_mds_md_v1 *)info->lti_ea_store;
537 lmm1->lmv_magic = cpu_to_le32(LMV_MAGIC);
538 lmm1->lmv_stripe_count = cpu_to_le32(stripe_count);
539 lmm1->lmv_hash_type = cpu_to_le32(lo->ldo_dir_hash_type);
540 rc = lod_fld_lookup(env, lod, lu_object_fid(&dt->do_lu),
545 lmm1->lmv_master_mdt_index = cpu_to_le32(mdtidx);
546 fid_cpu_to_le(&lmm1->lmv_stripe_fids[0], lu_object_fid(&dt->do_lu));
547 for (i = 0; i < lo->ldo_stripenr; i++) {
548 struct dt_object *dto;
550 dto = lo->ldo_stripe[i];
551 LASSERT(dto != NULL);
552 fid_cpu_to_le(&lmm1->lmv_stripe_fids[i + 1],
553 lu_object_fid(&dto->do_lu));
556 lmv_buf->lb_buf = info->lti_ea_store;
557 lmv_buf->lb_len = lmm_size;
558 lo->ldo_dir_striping_cached = 1;
563 int lod_parse_dir_striping(const struct lu_env *env, struct lod_object *lo,
564 const struct lu_buf *buf)
566 struct lod_thread_info *info = lod_env_info(env);
567 struct lod_device *lod = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
568 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
569 struct dt_object **stripe;
570 union lmv_mds_md *lmm = buf->lb_buf;
571 struct lmv_mds_md_v1 *lmv1 = &lmm->lmv_md_v1;
572 struct lu_fid *fid = &info->lti_fid;
577 if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
580 if (le32_to_cpu(lmv1->lmv_stripe_count) <= 1)
583 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[0]);
584 /* Do not load striping information for slave inode */
585 if (!lu_fid_eq(fid, lu_object_fid(&lo->ldo_obj.do_lu))) {
586 lo->ldo_dir_slave_stripe = 1;
590 LASSERT(lo->ldo_stripe == NULL);
591 OBD_ALLOC(stripe, sizeof(stripe[0]) *
592 (le32_to_cpu(lmv1->lmv_stripe_count) - 1));
596 /* skip master stripe */
597 for (i = 1; i < le32_to_cpu(lmv1->lmv_stripe_count); i++) {
598 struct lod_tgt_desc *tgt;
600 int type = LU_SEQ_RANGE_ANY;
601 struct dt_object *dto;
603 fid_le_to_cpu(fid, &lmv1->lmv_stripe_fids[i]);
604 rc = lod_fld_lookup(env, lod, fid, &idx, &type);
608 tgt = LTD_TGT(ltd, idx);
610 GOTO(out, rc = -ESTALE);
612 dto = dt_locate_at(env, tgt->ltd_tgt, fid,
613 lo->ldo_obj.do_lu.lo_dev->ld_site->ls_top_dev,
616 GOTO(out, rc = PTR_ERR(dto));
621 lo->ldo_stripe = stripe;
622 lo->ldo_stripenr = le32_to_cpu(lmv1->lmv_stripe_count) - 1;
623 lo->ldo_stripes_allocated = le32_to_cpu(lmv1->lmv_stripe_count) - 1;
625 lod_object_free_striping(env, lo);
630 static int lod_prep_md_striped_create(const struct lu_env *env,
631 struct dt_object *dt,
632 struct lu_attr *attr,
633 const struct lmv_user_md_v1 *lum,
636 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
637 struct lod_tgt_descs *ltd = &lod->lod_mdt_descs;
638 struct lod_object *lo = lod_dt_obj(dt);
639 struct dt_object **stripe;
640 struct lu_buf lmv_buf;
648 /* The lum has been verifed in lod_verify_md_striping */
649 LASSERT(le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC);
650 LASSERT(le32_to_cpu(lum->lum_stripe_count) > 0);
652 /* Do not need allocated master stripe */
653 stripe_count = le32_to_cpu(lum->lum_stripe_count);
654 OBD_ALLOC(stripe, sizeof(stripe[0]) * (stripe_count - 1));
658 OBD_ALLOC(idx_array, sizeof(idx_array[0]) * stripe_count);
659 if (idx_array == NULL)
660 GOTO(out_free, rc = -ENOMEM);
662 idx_array[0] = le32_to_cpu(lum->lum_stripe_offset);
663 for (i = 1; i < stripe_count; i++) {
664 struct lod_tgt_desc *tgt;
665 struct dt_object *dto;
668 struct lu_object_conf conf = { 0 };
670 idx = (idx_array[i - 1] + 1) % (lod->lod_remote_mdt_count + 1);
672 for (j = 0; j < lod->lod_remote_mdt_count;
673 j++, idx = (idx + 1) % (lod->lod_remote_mdt_count + 1)) {
674 bool already_allocated = false;
677 CDEBUG(D_INFO, "try idx %d, mdt cnt %d,"
678 " allocated %d, last allocated %d\n", idx,
679 lod->lod_remote_mdt_count, i, idx_array[i - 1]);
681 /* Find next avaible target */
682 if (!cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx))
685 /* check whether the idx already exists
686 * in current allocated array */
687 for (k = 0; k < i; k++) {
688 if (idx_array[k] == idx) {
689 already_allocated = true;
694 if (already_allocated)
700 /* Can not allocate more stripes */
701 if (j == lod->lod_remote_mdt_count) {
702 CDEBUG(D_INFO, "%s: require stripes %d only get %d\n",
703 lod2obd(lod)->obd_name, stripe_count, i - 1);
707 CDEBUG(D_INFO, "idx %d, mdt cnt %d,"
708 " allocated %d, last allocated %d\n", idx,
709 lod->lod_remote_mdt_count, i, idx_array[i - 1]);
711 tgt = LTD_TGT(ltd, idx);
712 LASSERT(tgt != NULL);
714 rc = obd_fid_alloc(tgt->ltd_exp, &fid, NULL);
719 conf.loc_flags = LOC_F_NEW;
720 dto = dt_locate_at(env, tgt->ltd_tgt, &fid,
721 dt->do_lu.lo_dev->ld_site->ls_top_dev, &conf);
723 GOTO(out_put, rc = PTR_ERR(dto));
728 lo->ldo_dir_striped = 1;
729 lo->ldo_stripe = stripe;
730 lo->ldo_stripenr = i - 1;
731 lo->ldo_stripes_allocated = stripe_count - 1;
733 if (lo->ldo_stripenr == 0)
734 GOTO(out_put, rc = -ENOSPC);
736 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
740 for (i = 0; i < lo->ldo_stripenr; i++) {
741 struct dt_object *dto;
744 /* only create slave striped object */
745 rc = dt_declare_create(env, dto, attr, NULL, NULL, th);
749 if (!dt_try_as_dir(env, dto))
750 GOTO(out_put, rc = -EINVAL);
752 rc = dt_declare_insert(env, dto,
753 (const struct dt_rec *)lu_object_fid(&dto->do_lu),
754 (const struct dt_key *)dot, th);
758 /* master stripe FID will be put to .. */
759 rc = dt_declare_insert(env, dto,
760 (const struct dt_rec *)lu_object_fid(&dt->do_lu),
761 (const struct dt_key *)dotdot, th);
765 /* probably nothing to inherite */
766 if (lo->ldo_striping_cached &&
767 !LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size,
768 lo->ldo_def_stripenr,
769 lo->ldo_def_stripe_offset)) {
770 struct lod_thread_info *info;
771 struct lov_user_md_v3 *v3;
773 /* sigh, lti_ea_store has been used for lmv_buf,
774 * so we have to allocate buffer for default
778 GOTO(out_put, rc = -ENOMEM);
780 memset(v3, 0, sizeof(*v3));
781 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
782 v3->lmm_stripe_count =
783 cpu_to_le32(lo->ldo_def_stripenr);
784 v3->lmm_stripe_offset =
785 cpu_to_le32(lo->ldo_def_stripe_offset);
786 v3->lmm_stripe_size =
787 cpu_to_le32(lo->ldo_def_stripe_size);
789 strncpy(v3->lmm_pool_name, lo->ldo_pool,
792 info = lod_env_info(env);
793 info->lti_buf.lb_buf = v3;
794 info->lti_buf.lb_len = sizeof(*v3);
795 rc = dt_declare_xattr_set(env, dto,
803 rc = dt_declare_xattr_set(env, dto, &lmv_buf, XATTR_NAME_LMV, 0,
809 rc = dt_declare_xattr_set(env, dt, &lmv_buf, XATTR_NAME_LMV, 0, th);
815 for (i = 0; i < stripe_count - 1; i++)
816 if (stripe[i] != NULL)
817 lu_object_put(env, &stripe[i]->do_lu);
818 OBD_FREE(stripe, sizeof(stripe[0]) * (stripe_count - 1));
822 if (idx_array != NULL)
823 OBD_FREE(idx_array, sizeof(idx_array[0]) * stripe_count);
829 * Declare create striped md object.
831 static int lod_declare_xattr_set_lmv(const struct lu_env *env,
832 struct dt_object *dt,
833 struct lu_attr *attr,
834 const struct lu_buf *lum_buf,
837 struct lod_object *lo = lod_dt_obj(dt);
838 struct lod_device *lod = lu2lod_dev(dt->do_lu.lo_dev);
839 struct lmv_user_md_v1 *lum;
843 lum = lum_buf->lb_buf;
844 LASSERT(lum != NULL);
846 CDEBUG(D_INFO, "lum magic = %x count = %u offset = %d\n",
847 le32_to_cpu(lum->lum_magic), le32_to_cpu(lum->lum_stripe_count),
848 (int)le32_to_cpu(lum->lum_stripe_offset));
850 if (le32_to_cpu(lum->lum_stripe_count) <= 1)
853 rc = lod_verify_md_striping(lod, lum);
857 /* prepare dir striped objects */
858 rc = lod_prep_md_striped_create(env, dt, attr, lum, th);
860 /* failed to create striping, let's reset
861 * config so that others don't get confused */
862 lod_object_free_striping(env, lo);
870 * LOV xattr is a storage for striping, and LOD owns this xattr.
871 * but LOD allows others to control striping to some extent
873 * - to set new defined striping
874 * - to set new semi-defined striping
875 * - number of stripes is defined
876 * - number of stripes + osts are defined
879 static int lod_declare_xattr_set(const struct lu_env *env,
880 struct dt_object *dt,
881 const struct lu_buf *buf,
882 const char *name, int fl,
885 struct dt_object *next = dt_object_child(dt);
886 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
892 * allow to declare predefined striping on a new (!mode) object
893 * which is supposed to be replay of regular file creation
894 * (when LOV setting is declared)
895 * LU_XATTR_REPLACE is set to indicate a layout swap
897 mode = dt->do_lu.lo_header->loh_attr & S_IFMT;
898 if ((S_ISREG(mode) || mode == 0) && strcmp(name, XATTR_NAME_LOV) == 0 &&
899 !(fl & LU_XATTR_REPLACE)) {
901 * this is a request to manipulate object's striping
903 if (dt_object_exists(dt)) {
904 rc = dt_attr_get(env, next, attr, BYPASS_CAPA);
908 memset(attr, 0, sizeof(*attr));
909 attr->la_valid = LA_TYPE | LA_MODE;
910 attr->la_mode = S_IFREG;
912 rc = lod_declare_striped_object(env, dt, attr, buf, th);
913 } else if (S_ISDIR(mode)) {
914 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
915 struct lod_object *lo = lod_dt_obj(dt);
918 if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
919 struct lmv_user_md_v1 *lum;
921 LASSERT(buf != NULL && buf->lb_buf != NULL);
923 rc = lod_verify_md_striping(d, lum);
928 rc = dt_declare_xattr_set(env, next, buf, name, fl, th);
932 /* set xattr to each stripes, if needed */
933 rc = lod_load_striping(env, lo);
937 if (lo->ldo_stripenr == 0)
940 for (i = 0; i < lo->ldo_stripenr; i++) {
941 LASSERT(lo->ldo_stripe[i]);
942 rc = dt_declare_xattr_set(env, lo->ldo_stripe[i], buf,
948 rc = dt_declare_xattr_set(env, next, buf, name, fl, th);
954 static void lod_lov_stripe_cache_clear(struct lod_object *lo)
956 lo->ldo_striping_cached = 0;
957 lo->ldo_def_striping_set = 0;
958 lod_object_set_pool(lo, NULL);
959 lo->ldo_def_stripe_size = 0;
960 lo->ldo_def_stripenr = 0;
963 static int lod_xattr_set_lov_on_dir(const struct lu_env *env,
964 struct dt_object *dt,
965 const struct lu_buf *buf,
966 const char *name, int fl,
968 struct lustre_capa *capa)
970 struct lod_device *d = lu2lod_dev(dt->do_lu.lo_dev);
971 struct dt_object *next = dt_object_child(dt);
972 struct lod_object *l = lod_dt_obj(dt);
973 struct lov_user_md_v1 *lum;
974 struct lov_user_md_v3 *v3 = NULL;
978 /* If it is striped dir, we should clear the stripe cache for
979 * slave stripe as well, but there are no effective way to
980 * notify the LOD on the slave MDT, so we do not cache stripe
981 * information for slave stripe for now. XXX*/
982 lod_lov_stripe_cache_clear(l);
983 LASSERT(buf != NULL && buf->lb_buf != NULL);
986 rc = lod_verify_striping(d, buf, 0);
990 if (lum->lmm_magic == LOV_USER_MAGIC_V3)
993 /* if { size, offset, count } = { 0, -1, 0 } and no pool
994 * (i.e. all default values specified) then delete default
995 * striping from dir. */
997 "set default striping: sz %u # %u offset %d %s %s\n",
998 (unsigned)lum->lmm_stripe_size,
999 (unsigned)lum->lmm_stripe_count,
1000 (int)lum->lmm_stripe_offset,
1001 v3 ? "from" : "", v3 ? v3->lmm_pool_name : "");
1003 if (LOVEA_DELETE_VALUES((lum->lmm_stripe_size),
1004 (lum->lmm_stripe_count),
1005 (lum->lmm_stripe_offset)) &&
1006 lum->lmm_magic == LOV_USER_MAGIC_V1) {
1007 rc = dt_xattr_del(env, next, name, th, capa);
1011 rc = dt_xattr_set(env, next, buf, name, fl, th, capa);
1017 static int lod_xattr_set_default_lmv_on_dir(const struct lu_env *env,
1018 struct dt_object *dt,
1019 const struct lu_buf *buf,
1020 const char *name, int fl,
1022 struct lustre_capa *capa)
1024 struct dt_object *next = dt_object_child(dt);
1025 struct lod_object *l = lod_dt_obj(dt);
1026 struct lmv_user_md_v1 *lum;
1030 LASSERT(buf != NULL && buf->lb_buf != NULL);
1033 CDEBUG(D_OTHER, "set default stripe_count # %u stripe_offset %d\n",
1034 le32_to_cpu(lum->lum_stripe_count),
1035 (int)le32_to_cpu(lum->lum_stripe_offset));
1037 if (LMVEA_DELETE_VALUES((le32_to_cpu(lum->lum_stripe_count)),
1038 le32_to_cpu(lum->lum_stripe_offset)) &&
1039 le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
1040 rc = dt_xattr_del(env, next, name, th, capa);
1044 rc = dt_xattr_set(env, next, buf, name, fl, th, capa);
1048 /* Update default stripe cache */
1049 if (l->ldo_dir_stripe == NULL) {
1050 OBD_ALLOC_PTR(l->ldo_dir_stripe);
1051 if (l->ldo_dir_stripe == NULL)
1055 l->ldo_dir_striping_cached = 0;
1056 l->ldo_dir_def_striping_set = 1;
1057 l->ldo_dir_def_stripenr =
1058 le32_to_cpu(lum->lum_stripe_count) - 1;
1064 static int lod_xattr_set_lmv(const struct lu_env *env, struct dt_object *dt,
1065 const struct lu_buf *buf, const char *name,
1066 int fl, struct thandle *th,
1067 struct lustre_capa *capa)
1069 struct lod_object *lo = lod_dt_obj(dt);
1070 struct lu_buf lmv_buf;
1075 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
1078 /* The stripes are supposed to be allocated in declare phase,
1079 * if there are no stripes being allocated, it will skip */
1080 if (lo->ldo_stripenr == 0)
1083 rc = lod_prep_lmv_md(env, dt, &lmv_buf);
1087 for (i = 0; i < lo->ldo_stripenr; i++) {
1088 struct dt_object *dto;
1089 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
1091 dto = lo->ldo_stripe[i];
1092 memset(attr, 0, sizeof(*attr));
1093 attr->la_valid = LA_TYPE | LA_MODE;
1094 attr->la_mode = S_IFDIR;
1095 rc = dt_create(env, dto, attr, NULL, NULL, th);
1099 rc = dt_insert(env, dto,
1100 (const struct dt_rec *)lu_object_fid(&dto->do_lu),
1101 (const struct dt_key *)dot, th, capa, 0);
1105 rc = dt_insert(env, dto,
1106 (struct dt_rec *)lu_object_fid(&dt->do_lu),
1107 (const struct dt_key *)dotdot, th, capa, 0);
1111 if (lo->ldo_striping_cached &&
1112 !LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size,
1113 lo->ldo_def_stripenr,
1114 lo->ldo_def_stripe_offset)) {
1115 struct lod_thread_info *info;
1116 struct lov_user_md_v3 *v3;
1118 /* sigh, lti_ea_store has been used for lmv_buf,
1119 * so we have to allocate buffer for default
1125 memset(v3, 0, sizeof(*v3));
1126 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
1127 v3->lmm_stripe_count =
1128 cpu_to_le32(lo->ldo_def_stripenr);
1129 v3->lmm_stripe_offset =
1130 cpu_to_le32(lo->ldo_def_stripe_offset);
1131 v3->lmm_stripe_size =
1132 cpu_to_le32(lo->ldo_def_stripe_size);
1134 strncpy(v3->lmm_pool_name, lo->ldo_pool,
1137 info = lod_env_info(env);
1138 info->lti_buf.lb_buf = v3;
1139 info->lti_buf.lb_len = sizeof(*v3);
1140 rc = dt_xattr_set(env, dto, &info->lti_buf,
1141 XATTR_NAME_LOV, 0, th, capa);
1147 rc = dt_xattr_set(env, dto, &lmv_buf, XATTR_NAME_LMV, fl, th,
1151 rc = dt_xattr_set(env, dt, &lmv_buf, XATTR_NAME_LMV, fl, th, capa);
1156 static int lod_xattr_set(const struct lu_env *env,
1157 struct dt_object *dt, const struct lu_buf *buf,
1158 const char *name, int fl, struct thandle *th,
1159 struct lustre_capa *capa)
1161 struct lod_object *lo = lod_dt_obj(dt);
1162 struct dt_object *next = dt_object_child(dt);
1168 attr = dt->do_lu.lo_header->loh_attr & S_IFMT;
1169 if (S_ISDIR(attr) && strcmp(name, XATTR_NAME_LOV) == 0) {
1170 rc = lod_xattr_set_lov_on_dir(env, dt, buf, name, fl, th, capa);
1171 } else if (S_ISREG(attr) && !strcmp(name, XATTR_NAME_LOV)) {
1172 /* in case of lov EA swap, just set it
1173 * if not, it is a replay so check striping match what we
1174 * already have during req replay, declare_xattr_set()
1175 * defines striping, then create() does the work
1177 if (fl & LU_XATTR_REPLACE) {
1178 /* free stripes, then update disk */
1179 lod_object_free_striping(env, lod_dt_obj(dt));
1180 rc = dt_xattr_set(env, next, buf, name, fl, th, capa);
1182 rc = lod_striping_create(env, dt, NULL, NULL, th);
1184 } else if (strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0) {
1187 rc = lod_xattr_set_default_lmv_on_dir(env, dt, buf, name, fl,
1191 * behave transparantly for all other EAs
1193 rc = dt_xattr_set(env, next, buf, name, fl, th, capa);
1196 if (rc != 0 || !S_ISDIR(attr))
1199 if (lo->ldo_stripenr == 0)
1202 for (i = 0; i < lo->ldo_stripenr; i++) {
1203 LASSERT(lo->ldo_stripe[i]);
1204 rc = dt_xattr_set(env, lo->ldo_stripe[i], buf, name, fl, th,
1213 static int lod_declare_xattr_del(const struct lu_env *env,
1214 struct dt_object *dt, const char *name,
1217 return dt_declare_xattr_del(env, dt_object_child(dt), name, th);
1220 static int lod_xattr_del(const struct lu_env *env, struct dt_object *dt,
1221 const char *name, struct thandle *th,
1222 struct lustre_capa *capa)
1224 if (!strcmp(name, XATTR_NAME_LOV))
1225 lod_object_free_striping(env, lod_dt_obj(dt));
1226 return dt_xattr_del(env, dt_object_child(dt), name, th, capa);
1229 static int lod_xattr_list(const struct lu_env *env,
1230 struct dt_object *dt, struct lu_buf *buf,
1231 struct lustre_capa *capa)
1233 return dt_xattr_list(env, dt_object_child(dt), buf, capa);
1236 int lod_object_set_pool(struct lod_object *o, char *pool)
1241 len = strlen(o->ldo_pool);
1242 OBD_FREE(o->ldo_pool, len + 1);
1247 OBD_ALLOC(o->ldo_pool, len + 1);
1248 if (o->ldo_pool == NULL)
1250 strcpy(o->ldo_pool, pool);
1255 static inline int lod_object_will_be_striped(int is_reg, const struct lu_fid *fid)
1257 return (is_reg && fid_seq(fid) != FID_SEQ_LOCAL_FILE);
1261 static int lod_cache_parent_lov_striping(const struct lu_env *env,
1262 struct lod_object *lp)
1264 struct lod_thread_info *info = lod_env_info(env);
1265 struct lov_user_md_v1 *v1 = NULL;
1266 struct lov_user_md_v3 *v3 = NULL;
1270 /* called from MDD without parent being write locked,
1272 dt_write_lock(env, dt_object_child(&lp->ldo_obj), 0);
1273 rc = lod_get_lov_ea(env, lp);
1277 if (rc < sizeof(struct lov_user_md)) {
1278 /* don't lookup for non-existing or invalid striping */
1279 lp->ldo_def_striping_set = 0;
1280 lp->ldo_striping_cached = 1;
1281 lp->ldo_def_stripe_size = 0;
1282 lp->ldo_def_stripenr = 0;
1283 lp->ldo_def_stripe_offset = (typeof(v1->lmm_stripe_offset))(-1);
1284 GOTO(unlock, rc = 0);
1288 v1 = info->lti_ea_store;
1289 if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1))
1290 lustre_swab_lov_user_md_v1(v1);
1291 else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3))
1292 lustre_swab_lov_user_md_v3(v3);
1294 if (v1->lmm_magic != LOV_MAGIC_V3 && v1->lmm_magic != LOV_MAGIC_V1)
1295 GOTO(unlock, rc = 0);
1297 if (v1->lmm_pattern != LOV_PATTERN_RAID0 && v1->lmm_pattern != 0)
1298 GOTO(unlock, rc = 0);
1300 lp->ldo_def_stripenr = v1->lmm_stripe_count;
1301 lp->ldo_def_stripe_size = v1->lmm_stripe_size;
1302 lp->ldo_def_stripe_offset = v1->lmm_stripe_offset;
1303 lp->ldo_striping_cached = 1;
1304 lp->ldo_def_striping_set = 1;
1305 if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
1306 /* XXX: sanity check here */
1307 v3 = (struct lov_user_md_v3 *) v1;
1308 if (v3->lmm_pool_name[0])
1309 lod_object_set_pool(lp, v3->lmm_pool_name);
1313 dt_write_unlock(env, dt_object_child(&lp->ldo_obj));
1318 static int lod_cache_parent_lmv_striping(const struct lu_env *env,
1319 struct lod_object *lp)
1321 struct lod_thread_info *info = lod_env_info(env);
1322 struct lmv_user_md_v1 *v1 = NULL;
1326 /* called from MDD without parent being write locked,
1328 dt_write_lock(env, dt_object_child(&lp->ldo_obj), 0);
1329 rc = lod_get_default_lmv_ea(env, lp);
1333 if (rc < sizeof(struct lmv_user_md)) {
1334 /* don't lookup for non-existing or invalid striping */
1335 lp->ldo_dir_def_striping_set = 0;
1336 lp->ldo_dir_striping_cached = 1;
1337 lp->ldo_dir_def_stripenr = 0;
1338 lp->ldo_dir_def_stripe_offset =
1339 (typeof(v1->lum_stripe_offset))(-1);
1340 lp->ldo_dir_def_hash_type = LMV_HASH_TYPE_FNV_1A_64;
1341 GOTO(unlock, rc = 0);
1345 v1 = info->lti_ea_store;
1347 lp->ldo_dir_def_stripenr = le32_to_cpu(v1->lum_stripe_count) - 1;
1348 lp->ldo_dir_def_stripe_offset = le32_to_cpu(v1->lum_stripe_offset);
1349 lp->ldo_dir_def_hash_type = le32_to_cpu(v1->lum_hash_type);
1350 lp->ldo_dir_def_striping_set = 1;
1351 lp->ldo_dir_striping_cached = 1;
1355 dt_write_unlock(env, dt_object_child(&lp->ldo_obj));
1359 static int lod_cache_parent_striping(const struct lu_env *env,
1360 struct lod_object *lp,
1366 rc = lod_load_striping(env, lp);
1370 if (!lp->ldo_striping_cached) {
1371 /* we haven't tried to get default striping for
1372 * the directory yet, let's cache it in the object */
1373 rc = lod_cache_parent_lov_striping(env, lp);
1378 if (S_ISDIR(child_mode) && !lp->ldo_dir_striping_cached)
1379 rc = lod_cache_parent_lmv_striping(env, lp);
1385 * used to transfer default striping data to the object being created
1387 static void lod_ah_init(const struct lu_env *env,
1388 struct dt_allocation_hint *ah,
1389 struct dt_object *parent,
1390 struct dt_object *child,
1393 struct lod_device *d = lu2lod_dev(child->do_lu.lo_dev);
1394 struct dt_object *nextp = NULL;
1395 struct dt_object *nextc;
1396 struct lod_object *lp = NULL;
1397 struct lod_object *lc;
1398 struct lov_desc *desc;
1403 if (likely(parent)) {
1404 nextp = dt_object_child(parent);
1405 lp = lod_dt_obj(parent);
1408 nextc = dt_object_child(child);
1409 lc = lod_dt_obj(child);
1411 LASSERT(lc->ldo_stripenr == 0);
1412 LASSERT(lc->ldo_stripe == NULL);
1415 * local object may want some hints
1416 * in case of late striping creation, ->ah_init()
1417 * can be called with local object existing
1419 if (!dt_object_exists(nextc) || dt_object_remote(nextc))
1420 nextc->do_ops->do_ah_init(env, ah, dt_object_remote(nextp) ?
1421 NULL : nextp, nextc, child_mode);
1423 if (S_ISDIR(child_mode)) {
1426 if (lc->ldo_dir_stripe == NULL) {
1427 OBD_ALLOC_PTR(lc->ldo_dir_stripe);
1428 if (lc->ldo_dir_stripe == NULL)
1432 if (lp->ldo_dir_stripe == NULL) {
1433 OBD_ALLOC_PTR(lp->ldo_dir_stripe);
1434 if (lp->ldo_dir_stripe == NULL)
1438 rc = lod_cache_parent_striping(env, lp, child_mode);
1442 /* transfer defaults to new directory */
1443 if (lp->ldo_striping_cached) {
1445 lod_object_set_pool(lc, lp->ldo_pool);
1446 lc->ldo_def_stripenr = lp->ldo_def_stripenr;
1447 lc->ldo_def_stripe_size = lp->ldo_def_stripe_size;
1448 lc->ldo_def_stripe_offset = lp->ldo_def_stripe_offset;
1449 lc->ldo_striping_cached = 1;
1450 lc->ldo_def_striping_set = 1;
1451 CDEBUG(D_OTHER, "inherite EA sz:%d off:%d nr:%d\n",
1452 (int)lc->ldo_def_stripe_size,
1453 (int)lc->ldo_def_stripe_offset,
1454 (int)lc->ldo_def_stripenr);
1457 /* transfer dir defaults to new directory */
1458 if (lp->ldo_dir_striping_cached) {
1459 lc->ldo_dir_def_stripenr = lp->ldo_dir_def_stripenr;
1460 lc->ldo_dir_def_stripe_offset =
1461 lp->ldo_dir_def_stripe_offset;
1462 lc->ldo_dir_def_hash_type =
1463 lp->ldo_dir_def_hash_type;
1464 lc->ldo_dir_striping_cached = 1;
1465 lc->ldo_dir_def_striping_set = 1;
1466 CDEBUG(D_INFO, "inherit default EA nr:%d off:%d t%u\n",
1467 (int)lc->ldo_dir_def_stripenr,
1468 (int)lc->ldo_dir_def_stripe_offset,
1469 lc->ldo_dir_def_hash_type);
1472 /* If the directory is specified with certain stripes */
1473 if (ah->dah_eadata != NULL && ah->dah_eadata_len != 0) {
1474 const struct lmv_user_md_v1 *lum1 = ah->dah_eadata;
1477 rc = lod_verify_md_striping(d, lum1);
1479 le32_to_cpu(lum1->lum_stripe_count) > 1) {
1480 /* Directory will be striped only if
1481 * stripe_count > 1 */
1483 le32_to_cpu(lum1->lum_stripe_count) - 1;
1484 lc->ldo_dir_stripe_offset =
1485 le32_to_cpu(lum1->lum_stripe_offset);
1486 lc->ldo_dir_hash_type =
1487 le32_to_cpu(lum1->lum_hash_type);
1488 CDEBUG(D_INFO, "set stripe EA nr:%hu off:%d\n",
1490 (int)lc->ldo_dir_stripe_offset);
1492 } else if (lp->ldo_dir_def_striping_set) {
1493 /* If there are default dir stripe from parent */
1494 lc->ldo_stripenr = lp->ldo_dir_def_stripenr;
1495 lc->ldo_dir_stripe_offset =
1496 lp->ldo_dir_def_stripe_offset;
1497 lc->ldo_dir_hash_type =
1498 lp->ldo_dir_def_hash_type;
1499 CDEBUG(D_INFO, "inherit EA nr:%hu off:%d\n",
1501 (int)lc->ldo_dir_stripe_offset);
1503 /* set default stripe for this directory */
1504 lc->ldo_stripenr = 0;
1505 lc->ldo_dir_stripe_offset = -1;
1508 CDEBUG(D_INFO, "final striping count:%hu, offset:%d\n",
1509 lc->ldo_stripenr, (int)lc->ldo_dir_stripe_offset);
1515 * if object is going to be striped over OSTs, transfer default
1516 * striping information to the child, so that we can use it
1517 * during declaration and creation
1519 if (!lod_object_will_be_striped(S_ISREG(child_mode),
1520 lu_object_fid(&child->do_lu)))
1523 * try from the parent
1525 if (likely(parent)) {
1526 lod_cache_parent_striping(env, lp, child_mode);
1528 lc->ldo_def_stripe_offset = (__u16) -1;
1530 if (lp->ldo_def_striping_set) {
1532 lod_object_set_pool(lc, lp->ldo_pool);
1533 lc->ldo_stripenr = lp->ldo_def_stripenr;
1534 lc->ldo_stripe_size = lp->ldo_def_stripe_size;
1535 lc->ldo_def_stripe_offset = lp->ldo_def_stripe_offset;
1536 CDEBUG(D_OTHER, "striping from parent: #%d, sz %d %s\n",
1537 lc->ldo_stripenr, lc->ldo_stripe_size,
1538 lp->ldo_pool ? lp->ldo_pool : "");
1543 * if the parent doesn't provide with specific pattern, grab fs-wide one
1545 desc = &d->lod_desc;
1546 if (lc->ldo_stripenr == 0)
1547 lc->ldo_stripenr = desc->ld_default_stripe_count;
1548 if (lc->ldo_stripe_size == 0)
1549 lc->ldo_stripe_size = desc->ld_default_stripe_size;
1550 CDEBUG(D_OTHER, "final striping: # %d stripes, sz %d from %s\n",
1551 lc->ldo_stripenr, lc->ldo_stripe_size,
1552 lc->ldo_pool ? lc->ldo_pool : "");
1555 /* we do not cache stripe information for slave stripe, see
1556 * lod_xattr_set_lov_on_dir */
1557 if (lp != NULL && lp->ldo_dir_slave_stripe)
1558 lod_lov_stripe_cache_clear(lp);
1563 #define ll_do_div64(aaa,bbb) do_div((aaa), (bbb))
1565 * this function handles a special case when truncate was done
1566 * on a stripeless object and now striping is being created
1567 * we can't lose that size, so we have to propagate it to newly
1570 static int lod_declare_init_size(const struct lu_env *env,
1571 struct dt_object *dt, struct thandle *th)
1573 struct dt_object *next = dt_object_child(dt);
1574 struct lod_object *lo = lod_dt_obj(dt);
1575 struct lu_attr *attr = &lod_env_info(env)->lti_attr;
1576 uint64_t size, offs;
1580 /* XXX: we support the simplest (RAID0) striping so far */
1581 LASSERT(lo->ldo_stripe || lo->ldo_stripenr == 0);
1582 LASSERT(lo->ldo_stripe_size > 0);
1584 rc = dt_attr_get(env, next, attr, BYPASS_CAPA);
1585 LASSERT(attr->la_valid & LA_SIZE);
1589 size = attr->la_size;
1593 /* ll_do_div64(a, b) returns a % b, and a = a / b */
1594 ll_do_div64(size, (__u64) lo->ldo_stripe_size);
1595 stripe = ll_do_div64(size, (__u64) lo->ldo_stripenr);
1597 size = size * lo->ldo_stripe_size;
1598 offs = attr->la_size;
1599 size += ll_do_div64(offs, lo->ldo_stripe_size);
1601 attr->la_valid = LA_SIZE;
1602 attr->la_size = size;
1604 rc = dt_declare_attr_set(env, lo->ldo_stripe[stripe], attr, th);
1610 * Create declaration of striped object
1612 int lod_declare_striped_object(const struct lu_env *env, struct dt_object *dt,
1613 struct lu_attr *attr,
1614 const struct lu_buf *lovea, struct thandle *th)
1616 struct lod_thread_info *info = lod_env_info(env);
1617 struct dt_object *next = dt_object_child(dt);
1618 struct lod_object *lo = lod_dt_obj(dt);
1622 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_ALLOC_OBDO)) {
1623 /* failed to create striping, let's reset
1624 * config so that others don't get confused */
1625 lod_object_free_striping(env, lo);
1626 GOTO(out, rc = -ENOMEM);
1629 /* choose OST and generate appropriate objects */
1630 rc = lod_qos_prep_create(env, lo, attr, lovea, th);
1632 /* failed to create striping, let's reset
1633 * config so that others don't get confused */
1634 lod_object_free_striping(env, lo);
1639 * declare storage for striping data
1641 info->lti_buf.lb_len = lov_mds_md_size(lo->ldo_stripenr,
1642 lo->ldo_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1);
1643 rc = dt_declare_xattr_set(env, next, &info->lti_buf, XATTR_NAME_LOV,
1649 * if striping is created with local object's size > 0,
1650 * we have to propagate this size to specific object
1651 * the case is possible only when local object was created previously
1653 if (dt_object_exists(next))
1654 rc = lod_declare_init_size(env, dt, th);
1660 int lod_dir_striping_create_internal(const struct lu_env *env,
1661 struct dt_object *dt,
1662 struct lu_attr *attr,
1663 const struct dt_object_format *dof,
1667 struct lod_thread_info *info = lod_env_info(env);
1668 struct dt_object *next = dt_object_child(dt);
1669 struct lod_object *lo = lod_dt_obj(dt);
1673 if (lo->ldo_dir_def_striping_set &&
1674 !LMVEA_DELETE_VALUES(lo->ldo_stripenr,
1675 lo->ldo_dir_stripe_offset)) {
1676 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
1677 int stripe_count = lo->ldo_stripenr + 1;
1679 if (info->lti_ea_store_size < sizeof(*v1)) {
1680 rc = lod_ea_store_resize(info, sizeof(*v1));
1683 v1 = info->lti_ea_store;
1686 memset(v1, 0, sizeof(*v1));
1687 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
1688 v1->lum_stripe_count = cpu_to_le32(stripe_count);
1689 v1->lum_stripe_offset =
1690 cpu_to_le32(lo->ldo_dir_stripe_offset);
1692 info->lti_buf.lb_buf = v1;
1693 info->lti_buf.lb_len = sizeof(*v1);
1696 rc = lod_declare_xattr_set_lmv(env, dt, attr,
1697 &info->lti_buf, th);
1699 rc = lod_xattr_set_lmv(env, dt, &info->lti_buf,
1700 XATTR_NAME_LMV, 0, th,
1706 /* Transfer default LMV striping from the parent */
1707 if (lo->ldo_dir_striping_cached &&
1708 !LMVEA_DELETE_VALUES(lo->ldo_dir_def_stripenr,
1709 lo->ldo_dir_def_stripe_offset)) {
1710 struct lmv_user_md_v1 *v1 = info->lti_ea_store;
1711 int def_stripe_count = lo->ldo_dir_def_stripenr + 1;
1713 if (info->lti_ea_store_size < sizeof(*v1)) {
1714 rc = lod_ea_store_resize(info, sizeof(*v1));
1717 v1 = info->lti_ea_store;
1720 memset(v1, 0, sizeof(*v1));
1721 v1->lum_magic = cpu_to_le32(LMV_USER_MAGIC);
1722 v1->lum_stripe_count = cpu_to_le32(def_stripe_count);
1723 v1->lum_stripe_offset =
1724 cpu_to_le32(lo->ldo_dir_def_stripe_offset);
1726 cpu_to_le32(lo->ldo_dir_def_hash_type);
1728 info->lti_buf.lb_buf = v1;
1729 info->lti_buf.lb_len = sizeof(*v1);
1731 rc = dt_declare_xattr_set(env, next, &info->lti_buf,
1732 XATTR_NAME_DEFAULT_LMV, 0,
1735 rc = dt_xattr_set(env, next, &info->lti_buf,
1736 XATTR_NAME_DEFAULT_LMV, 0, th,
1742 /* Transfer default LOV striping from the parent */
1743 if (lo->ldo_striping_cached &&
1744 !LOVEA_DELETE_VALUES(lo->ldo_def_stripe_size,
1745 lo->ldo_def_stripenr,
1746 lo->ldo_def_stripe_offset)) {
1747 struct lov_user_md_v3 *v3 = info->lti_ea_store;
1749 if (info->lti_ea_store_size < sizeof(*v3)) {
1750 rc = lod_ea_store_resize(info, sizeof(*v3));
1753 v3 = info->lti_ea_store;
1756 memset(v3, 0, sizeof(*v3));
1757 v3->lmm_magic = cpu_to_le32(LOV_USER_MAGIC_V3);
1758 v3->lmm_stripe_count = cpu_to_le16(lo->ldo_def_stripenr);
1759 v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset);
1760 v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size);
1762 strncpy(v3->lmm_pool_name, lo->ldo_pool,
1765 info->lti_buf.lb_buf = v3;
1766 info->lti_buf.lb_len = sizeof(*v3);
1769 rc = dt_declare_xattr_set(env, next, &info->lti_buf,
1770 XATTR_NAME_LOV, 0, th);
1772 rc = dt_xattr_set(env, next, &info->lti_buf,
1773 XATTR_NAME_LOV, 0, th,
1782 static int lod_declare_dir_striping_create(const struct lu_env *env,
1783 struct dt_object *dt,
1784 struct lu_attr *attr,
1785 struct dt_object_format *dof,
1788 return lod_dir_striping_create_internal(env, dt, attr, dof, th, true);
1791 static int lod_dir_striping_create(const struct lu_env *env,
1792 struct dt_object *dt,
1793 struct lu_attr *attr,
1794 struct dt_object_format *dof,
1797 return lod_dir_striping_create_internal(env, dt, attr, dof, th, false);
1800 static int lod_declare_object_create(const struct lu_env *env,
1801 struct dt_object *dt,
1802 struct lu_attr *attr,
1803 struct dt_allocation_hint *hint,
1804 struct dt_object_format *dof,
1807 struct dt_object *next = dt_object_child(dt);
1808 struct lod_object *lo = lod_dt_obj(dt);
1817 * first of all, we declare creation of local object
1819 rc = dt_declare_create(env, next, attr, hint, dof, th);
1823 if (dof->dof_type == DFT_SYM)
1824 dt->do_body_ops = &lod_body_lnk_ops;
1827 * it's lod_ah_init() who has decided the object will striped
1829 if (dof->dof_type == DFT_REGULAR) {
1830 /* callers don't want stripes */
1831 /* XXX: all tricky interactions with ->ah_make_hint() decided
1832 * to use striping, then ->declare_create() behaving differently
1833 * should be cleaned */
1834 if (dof->u.dof_reg.striped == 0)
1835 lo->ldo_stripenr = 0;
1836 if (lo->ldo_stripenr > 0)
1837 rc = lod_declare_striped_object(env, dt, attr,
1839 } else if (dof->dof_type == DFT_DIR) {
1840 rc = lod_declare_dir_striping_create(env, dt, attr, dof, th);
1846 int lod_striping_create(const struct lu_env *env, struct dt_object *dt,
1847 struct lu_attr *attr, struct dt_object_format *dof,
1850 struct lod_object *lo = lod_dt_obj(dt);
1854 LASSERT(lo->ldo_striping_cached == 0);
1856 /* create all underlying objects */
1857 for (i = 0; i < lo->ldo_stripenr; i++) {
1858 LASSERT(lo->ldo_stripe[i]);
1859 rc = dt_create(env, lo->ldo_stripe[i], attr, NULL, dof, th);
1865 rc = lod_generate_and_set_lovea(env, lo, th);
1870 static int lod_object_create(const struct lu_env *env, struct dt_object *dt,
1871 struct lu_attr *attr,
1872 struct dt_allocation_hint *hint,
1873 struct dt_object_format *dof, struct thandle *th)
1875 struct dt_object *next = dt_object_child(dt);
1876 struct lod_object *lo = lod_dt_obj(dt);
1880 /* create local object */
1881 rc = dt_create(env, next, attr, hint, dof, th);
1884 if (S_ISDIR(dt->do_lu.lo_header->loh_attr))
1885 rc = lod_dir_striping_create(env, dt, attr, dof, th);
1886 else if (lo->ldo_stripe && dof->u.dof_reg.striped != 0)
1887 rc = lod_striping_create(env, dt, attr, dof, th);
1893 static int lod_declare_object_destroy(const struct lu_env *env,
1894 struct dt_object *dt,
1897 struct dt_object *next = dt_object_child(dt);
1898 struct lod_object *lo = lod_dt_obj(dt);
1903 * we declare destroy for the local object
1905 rc = dt_declare_destroy(env, next, th);
1910 * load striping information, notice we don't do this when object
1911 * is being initialized as we don't need this information till
1912 * few specific cases like destroy, chown
1914 rc = lod_load_striping(env, lo);
1918 /* declare destroy for all underlying objects */
1919 for (i = 0; i < lo->ldo_stripenr; i++) {
1920 LASSERT(lo->ldo_stripe[i]);
1921 rc = dt_declare_destroy(env, lo->ldo_stripe[i], th);
1930 static int lod_object_destroy(const struct lu_env *env,
1931 struct dt_object *dt, struct thandle *th)
1933 struct dt_object *next = dt_object_child(dt);
1934 struct lod_object *lo = lod_dt_obj(dt);
1938 /* destroy local object */
1939 rc = dt_destroy(env, next, th);
1943 /* destroy all underlying objects */
1944 for (i = 0; i < lo->ldo_stripenr; i++) {
1945 LASSERT(lo->ldo_stripe[i]);
1946 /* for striped directory, next == ldo_stripe[0] */
1947 if (next != lo->ldo_stripe[i]) {
1948 rc = dt_destroy(env, lo->ldo_stripe[i], th);
1957 static int lod_index_try(const struct lu_env *env, struct dt_object *dt,
1958 const struct dt_index_features *feat)
1960 struct dt_object *next = dt_object_child(dt);
1964 LASSERT(next->do_ops);
1965 LASSERT(next->do_ops->do_index_try);
1967 rc = next->do_ops->do_index_try(env, next, feat);
1968 if (next->do_index_ops && dt->do_index_ops == NULL)
1969 dt->do_index_ops = &lod_index_ops;
1974 static int lod_declare_ref_add(const struct lu_env *env,
1975 struct dt_object *dt, struct thandle *th)
1977 return dt_declare_ref_add(env, dt_object_child(dt), th);
1980 static int lod_ref_add(const struct lu_env *env,
1981 struct dt_object *dt, struct thandle *th)
1983 return dt_ref_add(env, dt_object_child(dt), th);
1986 static int lod_declare_ref_del(const struct lu_env *env,
1987 struct dt_object *dt, struct thandle *th)
1989 return dt_declare_ref_del(env, dt_object_child(dt), th);
1992 static int lod_ref_del(const struct lu_env *env,
1993 struct dt_object *dt, struct thandle *th)
1995 return dt_ref_del(env, dt_object_child(dt), th);
1998 static struct obd_capa *lod_capa_get(const struct lu_env *env,
1999 struct dt_object *dt,
2000 struct lustre_capa *old, __u64 opc)
2002 return dt_capa_get(env, dt_object_child(dt), old, opc);
2005 static int lod_object_sync(const struct lu_env *env, struct dt_object *dt)
2007 return dt_object_sync(env, dt_object_child(dt));
2010 struct lod_slave_locks {
2012 struct lustre_handle lsl_handle[0];
2015 static int lod_object_unlock_internal(const struct lu_env *env,
2016 struct dt_object *dt,
2017 struct ldlm_enqueue_info *einfo,
2018 ldlm_policy_data_t *policy)
2020 struct lod_object *lo = lod_dt_obj(dt);
2021 struct lod_slave_locks *slave_locks = einfo->ei_cbdata;
2026 if (slave_locks == NULL)
2029 for (i = 0; i < slave_locks->lsl_lock_count; i++) {
2030 if (lustre_handle_is_used(&slave_locks->lsl_handle[i])) {
2033 einfo->ei_cbdata = &slave_locks->lsl_handle[i];
2034 rc1 = dt_object_unlock(env, lo->ldo_stripe[i], einfo,
2037 rc = rc == 0 ? rc1 : rc;
2044 static int lod_object_unlock(const struct lu_env *env, struct dt_object *dt,
2045 struct ldlm_enqueue_info *einfo,
2046 union ldlm_policy_data *policy)
2048 struct lod_object *lo = lod_dt_obj(dt);
2049 struct lod_slave_locks *slave_locks = einfo->ei_cbdata;
2050 int slave_locks_size;
2054 if (slave_locks == NULL)
2057 rc = lod_load_striping(env, lo);
2061 /* Note: for remote lock for single stripe dir, MDT will cancel
2062 * the lock by lockh directly */
2063 if (lo->ldo_stripenr == 0 && dt_object_remote(dt_object_child(dt)))
2066 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
2069 /* Only cancel slave lock for striped dir */
2070 rc = lod_object_unlock_internal(env, dt, einfo, policy);
2072 slave_locks_size = sizeof(*slave_locks) + slave_locks->lsl_lock_count *
2073 sizeof(slave_locks->lsl_handle[0]);
2074 OBD_FREE(slave_locks, slave_locks_size);
2075 einfo->ei_cbdata = NULL;
2080 static int lod_object_lock(const struct lu_env *env,
2081 struct dt_object *dt,
2082 struct lustre_handle *lh,
2083 struct ldlm_enqueue_info *einfo,
2084 union ldlm_policy_data *policy)
2086 struct lod_object *lo = lod_dt_obj(dt);
2089 int slave_locks_size;
2090 struct lod_slave_locks *slave_locks = NULL;
2093 /* remote object lock */
2094 if (!einfo->ei_enq_slave) {
2095 LASSERT(dt_object_remote(dt));
2096 return dt_object_lock(env, dt_object_child(dt), lh, einfo,
2100 if (!S_ISDIR(dt->do_lu.lo_header->loh_attr))
2103 rc = lod_load_striping(env, lo);
2108 if (lo->ldo_stripenr == 0)
2111 slave_locks_size = sizeof(*slave_locks) + lo->ldo_stripenr *
2112 sizeof(slave_locks->lsl_handle[0]);
2113 /* Freed in lod_object_unlock */
2114 OBD_ALLOC(slave_locks, slave_locks_size);
2115 if (slave_locks == NULL)
2117 slave_locks->lsl_lock_count = lo->ldo_stripenr;
2119 /* striped directory lock */
2120 for (i = 0; i < lo->ldo_stripenr; i++) {
2121 struct lustre_handle lockh;
2123 LASSERT(lo->ldo_stripe[i]);
2124 rc = dt_object_lock(env, lo->ldo_stripe[i], &lockh, einfo,
2129 slave_locks->lsl_handle[i] = lockh;
2132 einfo->ei_cbdata = slave_locks;
2135 if (rc != 0 && slave_locks != NULL) {
2136 einfo->ei_cbdata = slave_locks;
2137 lod_object_unlock_internal(env, dt, einfo, policy);
2138 OBD_FREE(slave_locks, slave_locks_size);
2139 einfo->ei_cbdata = NULL;
2145 struct dt_object_operations lod_obj_ops = {
2146 .do_read_lock = lod_object_read_lock,
2147 .do_write_lock = lod_object_write_lock,
2148 .do_read_unlock = lod_object_read_unlock,
2149 .do_write_unlock = lod_object_write_unlock,
2150 .do_write_locked = lod_object_write_locked,
2151 .do_attr_get = lod_attr_get,
2152 .do_declare_attr_set = lod_declare_attr_set,
2153 .do_attr_set = lod_attr_set,
2154 .do_xattr_get = lod_xattr_get,
2155 .do_declare_xattr_set = lod_declare_xattr_set,
2156 .do_xattr_set = lod_xattr_set,
2157 .do_declare_xattr_del = lod_declare_xattr_del,
2158 .do_xattr_del = lod_xattr_del,
2159 .do_xattr_list = lod_xattr_list,
2160 .do_ah_init = lod_ah_init,
2161 .do_declare_create = lod_declare_object_create,
2162 .do_create = lod_object_create,
2163 .do_declare_destroy = lod_declare_object_destroy,
2164 .do_destroy = lod_object_destroy,
2165 .do_index_try = lod_index_try,
2166 .do_declare_ref_add = lod_declare_ref_add,
2167 .do_ref_add = lod_ref_add,
2168 .do_declare_ref_del = lod_declare_ref_del,
2169 .do_ref_del = lod_ref_del,
2170 .do_capa_get = lod_capa_get,
2171 .do_object_sync = lod_object_sync,
2172 .do_object_lock = lod_object_lock,
2173 .do_object_unlock = lod_object_unlock,
2176 static ssize_t lod_read(const struct lu_env *env, struct dt_object *dt,
2177 struct lu_buf *buf, loff_t *pos,
2178 struct lustre_capa *capa)
2180 struct dt_object *next = dt_object_child(dt);
2181 return next->do_body_ops->dbo_read(env, next, buf, pos, capa);
2184 static ssize_t lod_declare_write(const struct lu_env *env,
2185 struct dt_object *dt,
2186 const loff_t size, loff_t pos,
2189 return dt_declare_record_write(env, dt_object_child(dt),
2193 static ssize_t lod_write(const struct lu_env *env, struct dt_object *dt,
2194 const struct lu_buf *buf, loff_t *pos,
2195 struct thandle *th, struct lustre_capa *capa, int iq)
2197 struct dt_object *next = dt_object_child(dt);
2199 return next->do_body_ops->dbo_write(env, next, buf, pos, th, capa, iq);
2202 static const struct dt_body_operations lod_body_lnk_ops = {
2203 .dbo_read = lod_read,
2204 .dbo_declare_write = lod_declare_write,
2205 .dbo_write = lod_write
2208 static int lod_object_init(const struct lu_env *env, struct lu_object *lo,
2209 const struct lu_object_conf *conf)
2211 struct lod_device *lod = lu2lod_dev(lo->lo_dev);
2212 struct lu_device *cdev = NULL;
2213 struct lu_object *cobj;
2214 struct lod_tgt_descs *ltd = NULL;
2215 struct lod_tgt_desc *tgt;
2217 int type = LU_SEQ_RANGE_ANY;
2221 rc = lod_fld_lookup(env, lod, lu_object_fid(lo), &idx, &type);
2225 if (type == LU_SEQ_RANGE_MDT &&
2226 idx == lu_site2seq(lo->lo_dev->ld_site)->ss_node_id) {
2227 cdev = &lod->lod_child->dd_lu_dev;
2228 } else if (type == LU_SEQ_RANGE_MDT) {
2229 ltd = &lod->lod_mdt_descs;
2231 } else if (type == LU_SEQ_RANGE_OST) {
2232 ltd = &lod->lod_ost_descs;
2239 if (ltd->ltd_tgts_size > idx &&
2240 cfs_bitmap_check(ltd->ltd_tgt_bitmap, idx)) {
2241 tgt = LTD_TGT(ltd, idx);
2243 LASSERT(tgt != NULL);
2244 LASSERT(tgt->ltd_tgt != NULL);
2246 cdev = &(tgt->ltd_tgt->dd_lu_dev);
2248 lod_putref(lod, ltd);
2251 if (unlikely(cdev == NULL))
2254 cobj = cdev->ld_ops->ldo_object_alloc(env, lo->lo_header, cdev);
2255 if (unlikely(cobj == NULL))
2258 lu_object_add(lo, cobj);
2263 void lod_object_free_striping(const struct lu_env *env, struct lod_object *lo)
2267 if (lo->ldo_dir_stripe != NULL) {
2268 OBD_FREE_PTR(lo->ldo_dir_stripe);
2269 lo->ldo_dir_stripe = NULL;
2272 if (lo->ldo_stripe) {
2273 LASSERT(lo->ldo_stripes_allocated > 0);
2275 for (i = 0; i < lo->ldo_stripenr; i++) {
2276 if (lo->ldo_stripe[i])
2277 lu_object_put(env, &lo->ldo_stripe[i]->do_lu);
2280 i = sizeof(struct dt_object *) * lo->ldo_stripes_allocated;
2281 OBD_FREE(lo->ldo_stripe, i);
2282 lo->ldo_stripe = NULL;
2283 lo->ldo_stripes_allocated = 0;
2285 lo->ldo_stripenr = 0;
2286 lo->ldo_pattern = 0;
2290 * ->start is called once all slices are initialized, including header's
2291 * cache for mode (object type). using the type we can initialize ops
2293 static int lod_object_start(const struct lu_env *env, struct lu_object *o)
2295 if (S_ISLNK(o->lo_header->loh_attr & S_IFMT))
2296 lu2lod_obj(o)->ldo_obj.do_body_ops = &lod_body_lnk_ops;
2300 static void lod_object_free(const struct lu_env *env, struct lu_object *o)
2302 struct lod_object *mo = lu2lod_obj(o);
2305 * release all underlying object pinned
2308 lod_object_free_striping(env, mo);
2310 lod_object_set_pool(mo, NULL);
2313 OBD_SLAB_FREE_PTR(mo, lod_object_kmem);
2316 static void lod_object_release(const struct lu_env *env, struct lu_object *o)
2318 /* XXX: shouldn't we release everything here in case if object
2319 * creation failed before? */
2322 static int lod_object_print(const struct lu_env *env, void *cookie,
2323 lu_printer_t p, const struct lu_object *l)
2325 struct lod_object *o = lu2lod_obj((struct lu_object *) l);
2327 return (*p)(env, cookie, LUSTRE_LOD_NAME"-object@%p", o);
2330 struct lu_object_operations lod_lu_obj_ops = {
2331 .loo_object_init = lod_object_init,
2332 .loo_object_start = lod_object_start,
2333 .loo_object_free = lod_object_free,
2334 .loo_object_release = lod_object_release,
2335 .loo_object_print = lod_object_print,